bmi and onsight survey

BMI and Climbing Ability

An overview of who has responded to the survey, so far.

Men

Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
tradn | 128 9.927734 1.353456 5 12.25
sportn | 134 10.9944 .9973962 7 13.25
boulder | 127 3.480315 1.617613 1 8
bmi | 138 23.26014 2.587718 17.2 32.7
age | 138 30.61594 8.882298 16 64

Women

Variable | Obs Mean Std. Dev. Min Max
-------------+--------------------------------------------------------
tradn | 7 8.107143 1.499008 6 10.75
sportn | 8 10.28125 .687094 9 11.25
boulder | 8 2.875 1.246423 1 5
bmi | 9 20.92222 2.662131 18.2 26.6
age | 9 32 10 24 56

Trad and sport are measured on the YDS scale, using just the decimal portion, and with letter qualifiers converted to 1/4 grade intervals. Bouldering is measured on the V scale.

In preliminary analyses, age has a marginally significant effect on trad onsight ability (positive). However, given the lack of a measurable effect on the other climbing outcomes, and the lack of a substantive effect on trad ability, I decided to drop age from the following analysis.

Gender appears to have a substantial effect (compare the means in each category). However, this particular group of women's data does not produce any measurable trend, so if we include them in the overall analysis we are just assuming their trend is the same as that of the men.

BMI and Trad

BMI does not explain a lot of the variation in trad onsight ability, but is explains some (R^2 is about 11%).

From these data, we would predict a person of average BMI (around 23) would climb about 5.10. One unit of change in BMI predicts about 0.18 full grade change in trad onsight ability.

(Men only. Note, BMI has been centered for the following analyses. A check for non-linear trend fails, see the discussion above.)

Source | SS df MS Number of obs = 128
-------------+------------------------------ F( 1, 126) = 16.59
Model | 27.0606023 1 27.0606023 Prob > F = 0.0001
Residual | 205.583441 126 1.63161461 R-squared = 0.1163
-------------+------------------------------ Adj R-squared = 0.1093
Total | 232.644043 127 1.83184286 Root MSE = 1.2773

------------------------------------------------------------------------------
tradn | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
bmic | -.1776534 .0436228 -4.07 0.000 -.2639817 -.0913252
_cons | 9.970376 .113387 87.93 0.000 9.745986 10.19477
------------------------------------------------------------------------------

BMI and trad ability, round 2

BMI and Sport

Results are similar to trad. BMI does a better job of predicting sport ability, with an R^2 of 19%. (So the correlation is stronger than for trad.)

The effect, however, is about the same. One unit change in BMI predicts about 0.17 full grade change in sport ability, but here the person of average BMI is predicted to be climbing about 5.11.

As I recall the numbers (see way above, not too well!) these results for sport and trad are consistent with the data Hans collected among more elite climbers, a few years back.

Source | SS df MS Number of obs = 134
-------------+------------------------------ F( 1, 132) = 31.74
Model | 25.6469199 1 25.6469199 Prob > F = 0.0000
Residual | 106.661382 132 .808040775 R-squared = 0.1938
-------------+------------------------------ Adj R-squared = 0.1877
Total | 132.308302 133 .994799265 Root MSE = .89891

------------------------------------------------------------------------------
sportn | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
bmic | -.1680803 .0298343 -5.63 0.000 -.2270955 -.1090651
_cons | 11.0224 .0778129 141.65 0.000 10.86848 11.17632
------------------------------------------------------------------------------

BMI and sport ability, round 2

BMI and Bouldering

I won't elaborate on these. Although the R^2 here has improved over the partial data, the changed coefficient (estimated effect of bmi) is still statistically the same. Keep in mind the response scale is the V scale, and not the YDS scale in the previous analyses.

Source | SS df MS Number of obs = 127
-------------+------------------------------ F( 1, 125) = 25.10
Model | 55.1418413 1 55.1418413 Prob > F = 0.0000
Residual | 274.558946 125 2.19647157 R-squared = 0.1672
-------------+------------------------------ Adj R-squared = 0.1606
Total | 329.700787 126 2.61667292 Root MSE = 1.482

------------------------------------------------------------------------------
boulder | Coef. Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
bmic | -.2608053 .0520521 -5.01 0.000 -.363823 -.1577877
_cons | 3.483272 .131512 26.49 0.000 3.222994 3.743551
------------------------------------------------------------------------------

BMI and bouldering ability, round 2

Optimistic wrote:I'd be happy to send the spreadsheet to anyone who wants to play with it... No argument from me that a better, more detailed study could've been conducted. For that matter, I've still got a couple weeks left on my month of SurveyMonkey if someone wants to design a few questions and start over! But I think that quantifying things like training and experience could turn out to be pretty hard to do in a meaningful way, and I'm assuming (perhaps wrongly?) that if we added in more variables, then we'd need more participants to achieve statistical significance, no? Maybe not...after all, this method of study is pretty cheap, I'm game to try again if folks have more ideas about how to do it better.

Not to be a dick but that's not how statistics work. You can't just leave out obvious confounding variables (like training/experience).

I mean, you can, but you just end up with a meaningless result.

Props for the effort and all though. It would be nice if there were more studies done on climbing.

"Essentially, all models are wrong, but some are useful." - George Box

If you find it meaningless, perhaps you expect too much.

Doug Hemken wrote:"Essentially, all models are wrong, but some are useful." - George Box If you find it meaningless, perhaps you expect too much.

Lol. Trolling?

You couldn't sneak this past the first day of Stats 101, let's be honest.

If you are just having fun with spreadsheets and surveys, that's cool. I'm not trying to rain on your parade or anything. Like I said, I think it's cool that you made the effort.

Of course it makes sense.

If you don't have a good BMI, it's way harder to send that day.

BMI, stands for Bowel Movement Index, right?

Ryan Watts wrote: Not to be a dick ....

Oh, I understand your intention.

Ryan Watts wrote: You can't just leave out obvious confounding variables (like training/experience).

Actually, you can!

For example, back in the '40's, essentially nothing was known about cancer. Then, people started doing epidemiologic studies and it started to become pretty clear that smoking had a very powerful association with lung cancer. They still had no idea what the causative agent was, and it remained true (as it does today) that a significant number of people who do not smoke get lung cancer. There are also other behaviors that bring additional risk of lung cancer, and these risks were not known at the time of the initial studies.

Your assertion, put another way, would be: you can't study anything prior to understanding it completely. This would be a pretty hard way to do science. How science actually does work: people ask a question, get some answers, and tell people the way they asked the question and what the answers were. Other people think about those results, and come up with a better way to ask the question.

If I picked up a journal article that said: "we have identified and controlled for ALL confounding variables", I would be highly suspicious that the paper was written by megalomaniacs. What actual scientists say is "this is what we think the confounders were, and these are the methods we used to deal with them. What do you think of our methods?"

Your comment also seems to imply that there can only be one independent variable. Most definitely there are many variables in climbing, and everyone here knows this. But demonstrating that experience is important would not prove that BMI is unimportant. You absolutely do not have to know all of the variables in play in order to study one variable. One simple example might be the mechanics of motion: you can be unaware of the concept of friction and yet still derive Newton's basic principles in the lab with a pretty decent degree of accuracy.

In this specific case, Doug has used some statistical tools to say that BMI appears to account for SOME of the variation seen in the data. This makes intuitive sense: we wouldn't have believed it if the data said that BMI accounted for all of the variation, and we wouldn't have believed it if the data said that BMI accounted for none of the variation. Obviously, there was very little in the way of rigor in this study: for all I know, ALL of the data may have been submitted by one person, making up numbers. However, I don't think folks would've been too eager to be submitting their social security number or some such for verification.

So we got what we got, and were transparent about how we did so. I think most of the grownups here understand that the data has some limitations (just like any scientific data), and will use it accordingly, or not!

Optimistic wrote: So we got what we got, and were transparent about how we did so. I think most of the grownups here understand that the data has some limitations (just like any scientific data), and will use it accordingly, or not!

I think that if you controlled for experience/training, BMI would have a much bigger correlation with performance, but you'd need a much larger data set. An R2 value of less than 0.2 is pretty poor.

Optimistic wrote: Actually, you can! For example, back in the '40's, essentially nothing was known about cancer. Then, people started doing epidemiologic studies and it started to become pretty clear that smoking had a very powerful association with lung cancer. They still had no idea what the causative agent was, and it remained true (as it does today) that a significant number of people who do not smoke get lung cancer. There are also other behaviors that bring additional risk of lung cancer, and these risks were not known at the time of the initial studies. Your assertion, put another way, would be: you can't study anything prior to understanding it completely. This would be a pretty hard way to do science. How science actually does work: people ask a question, get some answers, and tell people the way they asked the question and what the answers were. Other people think about those results, and come up with a better way to ask the question. If I picked up a journal article that said: "we have identified and controlled for ALL confounding variables", I would be highly suspicious that the paper was written by megalomaniacs. What actual scientists say is "this is what we think the confounders were, and these are the methods we used to deal with them. What do you think of our methods?" Your comment also seems to imply that there can only be one independent variable. Most definitely there are many variables in climbing, and everyone here knows this. But demonstrating that experience is important would not prove that BMI is unimportant. You absolutely do not have to know all of the variables in play in order to study one variable. One simple example might be the mechanics of motion: you can be unaware of the concept of friction and yet still derive Newton's basic principles in the lab with a pretty decent degree of accuracy. In this specific case, Doug has used some statistical tools to say that BMI appears to account for SOME of the variation seen in the data. This makes intuitive sense: we wouldn't have believed it if the data said that BMI accounted for all of the variation, and we wouldn't have believed it if the data said that BMI accounted for none of the variation. Obviously, there was very little in the way of rigor in this study: for all I know, ALL of the data may have been submitted by one person, making up numbers. However, I don't think folks would've been too eager to be submitting their social security number or some such for verification. So we got what we got, and were transparent about how we did so. I think most of the grownups here understand that the data has some limitations (just like any scientific data), and will use it accordingly, or not!

I'm just going to stop. I'm not teaching research methods 101 here.

If this thread has helped you improve your training/climbing or motivated you to lose weight or whatever, that's great.

MorganH wrote: I think that if you controlled for experience/training, BMI would have a much bigger correlation with performance, but you'd need a much larger data set. An R2 value of less than 0.2 is pretty poor.

Can you think of a good way to control for those things? Seems tricky...for example, I've been climbing for a large number of years, and I'm not anything like a good climber. Other people who've been climbing a lot less time than I have (and probably have higher BMI's in some cases!) climb a lot harder.

I wonder if what people are trying to get at is a controlled intervention: what happens when an individual climber loses or gains a lot of weight? But even this could be hard to do, because it's unlikely that a climber losing or gaining a lot of weight isn't also climbing a lot more or less at the same time, and so their ability may be changing for other reasons. So you'd have to have the subjects climbing the same amount and intensity before, during, and after the intervention.

Anyway, if someone can figure out how to ask these questions better, I'd be psyched to hear about it and see the results. Meantime, I'm on a diet. :)

Optimistic wrote: Can you think of a good way to control for those things? Seems tricky...for example, I've been climbing for a large number of years, and I'm not anything like a good climber. Other people who've been climbing a lot less time than I have (and probably have higher BMI's in some cases!) climb a lot harder. I wonder if what people are trying to get at is a controlled intervention: what happens when an individual climber loses or gains a lot of weight? But even this could be hard to do, because it's unlikely that a climber losing or gaining a lot of weight isn't also climbing a lot more or less at the same time, and so their ability may be changing for other reasons. So you'd have to have the subjects climbing the same amount and intensity before, during, and after the intervention. Anyway, if someone can figure out how to ask these questions better, I'd be psyched to hear about it and see the results. Meantime, I'm on a diet. :)

I'd say you are on the right track. The biggest problem (IMO) is that in the original, a third variable (training) is likely correlated with both BMI and ability. In theory, there could be NO causative relationship between BMI and ability, with training accounting for 100% of the correlation we see in the data.

Obviously you won't be able to eliminate all confounding variables like this but minimizing them to the point possible is crucial. A study where you took two roughly equal groups of climbers and had them follow the same training regimen with one group dieting to lose weight and the other group not dieting would have much more meaningful results. It's not bulletproof by any means but it would be a pretty reasonable setup.

Optimistic wrote: Can you think of a good way to control for those things? Seems tricky...for example, I've been climbing for a large number of years, and I'm not anything like a good climber. Other people who've been climbing a lot less time than I have (and probably have higher BMI's in some cases!) climb a lot harder. I wonder if what people are trying to get at is a controlled intervention: what happens when an individual climber loses or gains a lot of weight? But even this could be hard to do, because it's unlikely that a climber losing or gaining a lot of weight isn't also climbing a lot more or less at the same time, and so their ability may be changing for other reasons. So you'd have to have the subjects climbing the same amount and intensity before, during, and after the intervention. Anyway, if someone can figure out how to ask these questions better, I'd be psyched to hear about it and see the results. Meantime, I'm on a diet. :)

I think if you grouped the data by number of hours spent climbing/training per week and number of years spent climbing (basically, drop it into bins), you'd end up with a bunch of datasets that you could analyze individually. In order for it to be meaningful, you'd need a lot of responses though.

For instance:

Group 1: climbed 2 years or less, climb 10 hours/week or less
Group 2: climb 2 years or less, climb 10-20 hours/week
Group 3: climb 40 hours/week

Group 5: climbed 2-5 years, climb 10 hours/week or less
Group 6: climbed 2-5 years, climb 10-20 hours/week
Group 7: climbed 2-5 years, climb 10-30 hours/week
Group 8: climbed 2-5 years, climb > 40 hours/week

Group 9: climbed > 5 years, climb 10 hours/week or less
Group 10: climbed > 5 years, climb 10-20 hours/week
Group 11: climbed > 5 years, climb 10-30 hours/week
Group 12: climbed > 5 years, climb > 40 hours/week

If you could get 100 or so responses in each group, I bet you'd see a strong correlation (>0.5?) between BMI & max onsite/flash grades, whether it be sport, trad, or bouldering.

Why not just treat these as two new variables? Then a sample of 100-150 would probably be sufficient.

Personally, I doubt the partial R^2 for BMI would climb to .25 once you controlled for hours trained/week, if anything I would expect it to drop (as Ryan describes, but not as drastically as he hypothesizes). I think controlling for experience wouldn't have much effect on the partial R^2 between BMI and climbing ability. My opinion.

PM sent regarding the dataset. Thanks for putting this together, Optimistic!

The guys that wrote the Squamish guidebook have got it all figured out. This is for trad leads.

5.10 The happy level that most accomplished weekend warriors are able to lead.

5.11 Requires training for most people, or regular hard climbing. Steady work can feel like a drag.

5.12 Requires much time at the crag, training, and natural talent. Hard to achieve with gainful employment, but some do manage it.

5.13 Demands great talent and full dedication. Employment is seen, at best, as an irritant.

5.14 Well, keep trying.

drew abney wrote:PM sent regarding the dataset. Thanks for putting this together, Optimistic!

As Drew pointed out in his PM, data set will be shared, but stripped of all identifying information.

Have fun!
David

Myself, I think I've gotten what I needed from the data, which is a rough approximation of a target weight for myself, based on what others at the target grade I'd like to be climbing seem to weigh, on average.

And if I hit the weight target and nothing happens...well, then it's banana splits and Belgian Tripel Ale all around, I guess!